Service models and setting

TL;DR — Setting is a risk-management and treatment-intensity decision, not a proxy for diagnosis severity based on BMI alone. Outpatient specialist care is preferred when medical and psychiatric risk can be managed; inpatient care is for instability or when necessary treatment cannot be delivered safely. In the ANDI non-inferiority RCT, 172 female adolescents aged 11–18 — all below the tenth BMI percentile and all in their first hospital admission for AN — were randomized after three weeks of inpatient care to day-patient or continued inpatient treatment at six German centres. Day treatment was non-inferior on the primary outcome, BMI at 12-month follow-up (mean difference 0.46 kg/m² favouring day patient, 95% CI −0.11 to 1.02; p for non-inferiority <0.0001, margin 0.75 kg/m²), with a similar number of treatment-related serious adverse events (eight inpatient, seven day patient). The trial did not report a cost analysis; its authors conclude that day treatment "might be a safe and less costly alternative" (Herpertz-Dahlmann 2014, PMID 24439238). Continuity between settings is likely as important as the bed label.

Setting matrix

Setting	Principal function	Advantages	Risks/limits
Outpatient	Specialist psychotherapy, nutrition and monitoring	Maintains ordinary life; lower disruption	Insufficient for instability or uncontrolled loss
Intensive outpatient	Multiple contacts while living at home	More meal/behaviour support	Variable definition and availability
Day patient	Structured meals/therapy by day	Intensity with overnight home generalization	Requires safe nights and transport/support
Psychiatric inpatient	Containment, structured nutrition, psychiatric care	High observation/intensity	Institutionalization, coercion, transitions
Medical inpatient	Acute stabilization and refeeding	Organ monitoring and rapid intervention	Not a substitute for specialist ED therapy
Residential	24-hour non-hospital support in some systems	Longer behavioural milieu	Regulatory/evidence heterogeneity

ANDI trial

The multicentre randomized open-label non-inferiority trial compared day-patient treatment after short inpatient stabilization with continued inpatient treatment for adolescents. Its eligibility criteria carry most of the interpretive weight: female, aged 11–18, BMI below the tenth percentile, first admission, and already three weeks into inpatient care at randomization. The result therefore supports day care as an alternative for selected, already-stabilized, non-chronic patients in an experienced system. It does not establish equivalence for acutely unstable patients, for people with repeated admissions or long illness duration, for other health systems, for males, or for adults (Herpertz-Dahlmann 2014, PMID 24439238). Three of the inpatient-group and two of the day-patient-group serious adverse events related to suicidal ideation, and one day-patient participant attempted suicide three months after discharge — a reminder that psychiatric risk does not track the setting decision neatly.

Roughly one fifth to one third of adolescents with AN require intensive care at some point in their illness, so the setting question applies to a substantial minority rather than to everyone; a review spanning inpatient, day-patient and home treatment across health systems notes that the mid- to long-term outcomes and costs of these programmes have not been compared, and calls that comparison urgent (Herpertz-Dahlmann 2021, PMID 33924294).

What randomized setting evidence exists overall

Beyond ANDI, the randomized setting literature is thinner than the intensity of clinical debate suggests. The Cochrane review of treatment setting identified five eligible randomized trials, four of them in AN with 511 participants in total. For AN there may be little or no difference between specialist inpatient care and active outpatient or combined brief-hospital-plus-outpatient care in weight gain 12 months after treatment start (SMD −0.22, 95% CI −0.49 to 0.05; 2 trials, 232 participants; low-quality evidence), and people may be more likely to complete treatment when randomized to outpatient settings, though that estimate is rated very uncertain (RR 0.75, 95% CI 0.64 to 0.88; 3 trials, 319 participants; very-low-quality evidence). The reviewers concluded there was insufficient evidence that any setting is superior for moderately severe or less severe AN or other eating disorders (Hay 2019, PMID 30663033).

The corresponding adolescent trial of hospitalization duration reached the same place from a different direction: 82 medically unstable adolescents randomized to shorter hospitalization for medical stabilization or longer hospitalization for weight restoration to 90% expected body weight, each followed by 20 sessions of FBT, showed no significant difference in hospital days after the initial admission, while the weight-restoration arm used significantly more total hospital days (Madden 2015, PMID 25017941).

For adults the question is currently unanswerable rather than answered. DAISIES — a non-inferiority trial of inpatient treatment-as-usual versus stepped-care day-patient treatment in adults with BMI ≤ 16 kg/m² — intended 386 participants, approached 53 patients over 16 months and randomized 15 before terminating. Strong treatment preference among patients was the most common reason for non-participation, compounded by COVID-19 service disruption; both arms gained BMI similarly, but no effectiveness conclusion is possible. Patients and carers perceived the day-patient approach as more acceptable (İnce 2025, PMID 39943786). The next randomized attempt at a genuinely different model is the German HoT trial, which will randomize adolescents after 5–8 weeks of inpatient somatic stabilization to continued inpatient/day-patient care or multidisciplinary home treatment, with 12-month BMI as the primary outcome and rehospitalization, carer burden and direct treatment costs among secondary outcomes (Dahmen 2024, PMID 39538317; DRKS00025925). It had not reported as of September 2026.

The trial that tested the whole English pathway

TOuCAN is the only pragmatic trial to randomize the service level rather than a protocol, and its findings still shape UK practice. Across 35 English child and adolescent mental health services, 215 young people with AN were identified and 167 (mean age 14 years 11 months) randomized to inpatient treatment in one of four experienced units, a specialist outpatient programme, or treatment as usual in general community CAMHS, with a further 48 followed as a preference group.

Finding	Detail
Adherence	65% overall, and significantly lower for inpatient allocation (49%) than for general CAMHS (71%) or specialist outpatient (77%), p = 0.013
Effectiveness	Significant improvement in all groups at every timepoint; good outcome in 19% at 1 year, 33% at 2 years and 64% of those followed at 5 years. No difference between inpatient and outpatient, or specialist and generalist, once baseline characteristics were accounted for
Follow-up	Outcome measure completed for 94% at 1 year, 93% at 2 years and 47% at 5 years
Cost	Generalist CAMHS was slightly more expensive over 2 years, largely because more of those patients were subsequently admitted. The specialist outpatient programme was the dominant strategy on incremental cost-effectiveness
Acceptability	Parental satisfaction generally good and better with specialist care; young people's satisfaction much more mixed, again better with specialist care

(Gowers 2010, PMID 20334748; ISRCTN39345394.) The authors' conclusion is unusually direct: the study "provides little support for lengthy inpatient psychiatric treatment on clinical or health economic grounds", and specifically lends "little support to a stepped-care approach in which inpatient care is offered to outpatient non-responders". The design limitation is equally direct — 51% non-adherence in the inpatient arm means the inpatient effect is poorly estimated, and the same treatment-preference problem that later destroyed DAISIES recruitment (İnce 2025, PMID 39943786) is visible here as differential adherence.

Cost-effectiveness

Analysis	Comparison	Result
Le 2017, PMID 29044637	Markov model, 11–18-year-olds with short-duration AN, Australian health-system perspective, 6 years	FBT was less costly than adolescent-focused therapy. Against no intervention, mean ICER was AUD 5,089 per DALY averted for FBT (95% UI dominant to 16,659) and AUD 51,897 for AFT (21,591 to 1,712,491). At a AUD 50,000/DALY threshold, FBT was 100% likely and AFT 45% likely to be cost-effective
Mayr 2025, PMID 39776084	Non-randomized pilot, 61 adolescents: FBT + home treatment (n=40) vs FBT alone (n=21)	Weight restoration within three months 70% vs 52% at >85% expected body weight and 38% vs 29% at >95%; cost per patient meeting criteria CHF 8,243 vs 20,446 and CHF 15,387 vs 37,485. Inpatient treatment accounted for most of the cost, and home treatment reduced hospitalization
Guarda 2017, PMID 28130794	314 consecutive adult first admissions to a US inpatient/partial-hospitalization programme, 2003–2015	$2,295 per inpatient day and $1,567 per partial-hospitalization day; $4,089 and $7,050 per pound gained; 70% reached discharge BMI ≥19. Residential care compared unfavourably on both cost per pound and likelihood of weight restoration

Across these analyses, inpatient bed-days are a major cost driver. Reduced admissions can improve cost-effectiveness, but only if intervention costs and clinical outcomes are considered alongside them; none of these studies establishes that every community alternative will save money or preserve outcomes.

Community, home and residential care

A scoping review of intensive community treatment and home-based treatment for eating disorders — defined as more than two planned therapeutic contacts per week, excluding physical monitoring — found 46 sources (31 intensive community, 15 home-based), mostly from Europe (23) and the USA (18), of which 28 reported quantitative data. Home-based treatments mainly targeted children and adolescents with AN using family-based approaches; intensive community treatments varied more in age and diagnosis and often combined cognitive-behavioural and dialectical-behavioural elements. Studies frequently reported clinical improvement, feasibility, patient satisfaction, strong adherence and cost-effectiveness through reduced admissions — and there were no randomized controlled trials at all (İnce 2025, PMID 41214774).

Residential care, which dominates the private US intensive-treatment market, has a comparable evidence structure. A narrative review found 22 partial-hospitalization and 6 residential studies reporting discharge response (generally improvement), but only 9 and 3 respectively reporting post-discharge outcome, with substantial loss to follow-up that was particularly severe for residential treatment. Only two follow-up studies used controlled designs, both showing partial hospitalization superior to inpatient care for maintaining remission; the reviewers concluded that improvement at discharge may not predict long-term outcome and that controlled evidence of long-term efficacy for residential treatment is lacking (Friedman 2016, PMID 27062687).

Access and delay as a service outcome

Setting is only one dimension of service design; the other is how long people wait. Duration of untreated eating disorder is reported internationally as 2.5–6 years, and the FREED early-intervention model was built to compress it.

Evaluation	Design	Result
Brown 2018, PMID 27619198	60 recruited, 51 compared with 89 patients seen earlier in the same service	Shorter waits for assessment and treatment, better treatment uptake, high satisfaction. Reductions in duration of untreated illness depended on gatekeeping: with minimal gatekeeping, duration of untreated eating disorder 13.0 vs 19.1 months; with complex gatekeeping, 17.7 months and a longer time to specialist contact than controls
Flynn 2021, PMID 33112472	Multicentre quasi-experimental, 278 FREED patients aged 16–25 with <3 years' illness vs 224 historical treatment-as-usual	Significantly shorter duration of untreated eating disorder and waiting times; duration reduced by ~4 months where systemic delays were minimal; treatment uptake 97.8% vs 75.4%
Richards 2023, PMID 35676870	National English roll-out (FREED-4-All), 2,473 patients, Sept 2018–Sept 2021, benchmarked against the 278-patient study	Duration of untreated eating disorder 15 vs 18 months; wait-time target adherence comparable (~85% engagement calls attempted <2 days, ~50–60% assessments offered <14 days, ~40% treatment offered <28 days); symptom improvement comparable — but only 6% of patients had post-treatment data, and the authors themselves describe the data as "flawed, uncertain, proximate, and sparse"

Three points follow. The uptake difference (97.8% vs 75.4%) is the largest single number in the service literature and does not depend on the weakest part of the evidence. The gatekeeping finding shows that a specialist service can shorten its own queue while the total delay is set upstream of it. And the national roll-out demonstrates process replication with almost no outcome data — a pattern that would not be accepted for a drug and is routinely accepted for a service model. None of these evaluations is randomized.

Decision variables

Variable	Favors greater intensity
Physiology	Severe or worsening bradycardia/hypotension, electrolyte/glucose abnormality, dehydration, hypothermia, organ dysfunction
Trajectory	Rapid ongoing loss or inability to interrupt restriction
Refeeding	Need for monitoring beyond outpatient capacity
Psychiatric risk	Acute suicide/self-harm risk, severe comorbidity, inability to maintain safety
Environment	Insufficient meal support or safeguarding concern
Treatment response	Failure of a credible, delivered lower-intensity plan

NICE NG69 recommendation 1.11.2 states directly: do not use an absolute weight or BMI threshold when deciding whether to admit someone with an eating disorder to day-patient or inpatient care; recommendation 1.11.1 sets the positive trigger as physical health being severely compromised such that stabilisation and refeeding cannot be done as an outpatient. Rate of loss, medical parameters and the capacity of the proposed setting matter (NICE NG69, recommendation text retrieved and read 2026-09-02; Society for Adolescent Health and Medicine 2022, PMID 36058805).

NICE also names a harm on the other side of the decision, which few guidelines do: a person may become institutionalised by a long admission, a lack of change in their condition could indicate that inpatient treatment is harmful, and an independent second opinion should be considered where clinicians disagree about continued inpatient care (NICE NG69 §1.11, retrieved 2026-09-02). Setting escalation is therefore framed as a decision with risk in both directions rather than as a safety default.

Outcomes services should report

Weight at discharge is insufficient. Services should report medical stability, standardized weight trajectory, eating-disorder symptoms, functioning, patient-reported experience, restraint/compulsory feeding, unplanned transfer, readmission, follow-up retention and cost. Trial attrition is itself an outcome in adult AN, where 22% were lost by end of treatment and 30% by 12 months in the largest randomized trial (Zipfel 2014, PMID 24131861).

Readmission deserves particular emphasis because the inpatient literature's headline metrics conceal it. In the StRONG trial, a refeeding protocol that shortened hospital stay by 4.0 days produced no reduction in medical rehospitalization over the following year — roughly a third of participants in both arms were readmitted at least once (32.8% vs 35.4%) (Garber 2021, PMID 33074282; Golden 2021, PMID 33753542). A service reporting only length of stay would have recorded that protocol as a clear success and missed the fact that a third of its patients came back. Refeeding protocol design and setting choice are the same problem viewed from two angles (Garber 2016, PMID 26661289).

A large transition trial has tested a digital aftercare augmentation, though not with readmission as its primary endpoint. TRIANGLE randomized 371 patient-carer dyads after inpatient or day-patient care to ECHOMANTRA plus usual care or usual care alone. Patient distress at 12 months did not differ (p=0.87), secondary outcomes were also non-significant, and the intervention was more costly while producing fewer quality-adjusted life-years; only 20% met the minimum interactive-component adherence threshold (Cardi 2024, PMID 38841708). This is evidence against that specific digital package, not against structured transition support in general.

Open questions

• Which patients can transition safely to day care after brief stabilization? ANDI answers this for first-admission female adolescents already stabilized over three weeks. DAISIES did randomize adults, but stopped after 15 participants and could not answer effectiveness; as of September 2026 no adequately recruited adult trial, or randomized trial focused on males or people with repeated admissions, has answered the comparison (Herpertz-Dahlmann 2014, PMID 24439238; İnce 2025, PMID 39943786).
• Do intensive community models reduce compulsory treatment and readmission without worsening medical outcomes?
• Which transition interventions prevent the post-discharge loss of gains? Around a third of adolescents refed in a randomized inpatient protocol were medically rehospitalized within a year regardless of the protocol used. TRIANGLE tested one digital transition package in 371 dyads and was null on its distress primary outcome; readmission was not primary, so the rehospitalization question remains open (Golden 2021, PMID 33753542; Cardi 2024, PMID 38841708).
• Do mid- and long-term outcomes and costs actually differ between inpatient, day-patient and home-treatment models across health systems, as reviewers have repeatedly asked (Herpertz-Dahlmann 2021, PMID 33924294)?
• Does home treatment after brief stabilization outperform continued inpatient or day-patient care? The first randomized test is under way and has not reported (Dahmen 2024, PMID 39538317).
• Does early intervention change outcome, or only access? FREED reliably shortens delay and raises treatment uptake, but its largest evaluation had post-treatment data on 6% of patients and none of the evaluations is randomized (Flynn 2021, PMID 33112472; Richards 2023, PMID 35676870).
• How much of the duration of untreated illness is attributable to specialist services rather than to referral gatekeeping upstream of them (Brown 2018, PMID 27619198)?
• Can adult setting questions be answered by randomization at all, given that treatment preference was the leading cause of DAISIES' failure to recruit (İnce 2025, PMID 39943786; Hay 2019, PMID 30663033)?

Related pages

• Refeeding and nutritional rehabilitation — monitored restoration.
• Guidelines — setting recommendations.
• Severe and enduring illness and compulsory treatment — coercion in intensive care.

References

1. Herpertz-Dahlmann B, et al. Day-patient treatment after short inpatient care versus continued inpatient treatment in adolescents with anorexia nervosa. Lancet. 2014. PMID 24439238.
2. Society for Adolescent Health and Medicine. Medical management of restrictive eating disorders. J Adolesc Health. 2022. PMID 36058805.
3. Garber AK, et al. Short-term outcomes of higher- versus lower-calorie refeeding. JAMA Pediatr. 2021. PMID 33074282.
4. Herpertz-Dahlmann B, et al. Intensive treatments in adolescent anorexia nervosa. Nutrients. 2021. PMID 33924294.
5. Zipfel S, et al. ANTOP randomized controlled trial. Lancet. 2014. PMID 24131861.
6. Garber AK, et al. Systematic review of refeeding approaches. Int J Eat Disord. 2016. PMID 26661289.
7. Golden NH, et al. Higher-calorie refeeding in anorexia nervosa: 1-year outcomes from a randomized controlled trial. Pediatrics. 2021;147:e2020037135. PMID 33753542.
8. Hay PJ, et al. Inpatient versus outpatient care, partial hospitalisation and waiting list for people with eating disorders. Cochrane Database Syst Rev. 2019;1:CD010827. PMID 30663033.
9. Madden S, et al. A randomized controlled trial of in-patient treatment for anorexia nervosa in medically unstable adolescents. Psychol Med. 2015;45:415-427. PMID 25017941.
10. İnce B, et al. Stepping into day treatment approach versus inpatient treatment for adults with anorexia nervosa: the DAISIES RCT. Health Technol Assess. 2025;29:1-37. PMID 39943786.
11. Dahmen B, et al. Anorexia nervosa-specific home treatment in children and adolescents and their families (the HoT study): study protocol. Trials. 2024;25:760. Protocol. PMID 39538317. (DRKS00025925)
12. Brown A, et al. The FREED Project (First Episode and Rapid Early Intervention in Eating Disorders): service model, feasibility and acceptability. Early Interv Psychiatry. 2018;12:250-257. PMID 27619198.
13. Flynn M, et al. Assessing the impact of First Episode Rapid Early Intervention for Eating Disorders on duration of untreated eating disorder: a multi-centre quasi-experimental study. Eur Eat Disord Rev. 2021;29:458-471. PMID 33112472.
14. Richards KL, et al. National roll-out of early intervention for eating disorders: process and clinical outcomes from First Episode Rapid Early Intervention for Eating Disorders. Early Interv Psychiatry. 2023;17:202-211. PMID 35676870.
15. Gowers SG, et al. A randomised controlled multicentre trial of treatments for adolescent anorexia nervosa including assessment of cost-effectiveness and patient acceptability — the TOuCAN trial. Health Technol Assess. 2010;14:1-98. PMID 20334748. (ISRCTN39345394)
16. Le LK, et al. The modeled cost-effectiveness of family-based and adolescent-focused treatment for anorexia nervosa. Int J Eat Disord. 2017;50:1356-1366. PMID 29044637.
17. Mayr LM, et al. Cost-effectiveness analysis of family-based treatment with additional home treatment for adolescent anorexia nervosa. Eur Eat Disord Rev. 2025;33:608-615. PMID 39776084.
18. Guarda AS, et al. Intensive treatment for adults with anorexia nervosa: the cost of weight restoration. Int J Eat Disord. 2017;50:302-306. PMID 28130794.
19. İnce B, et al. Intensive community and home-based treatments for eating disorders: a scoping review. J Eat Disord. 2025;13:256. PMID 41214774.
20. Friedman K, et al. A narrative review of outcome studies for residential and partial hospital-based treatment of eating disorders. Eur Eat Disord Rev. 2016;24:263-276. PMID 27062687.
21. Cardi V, et al. Efficacy and cost-effectiveness of a digital guided self-management intervention to support transition from intensive care to community care in anorexia nervosa (TRIANGLE). EClinicalMedicine. 2024;73:102645. PMID 38841708.